Elon: "Feature complete for full self driving this year"

[electronblue]

But that is not more torque, it is just more internal torque checks/ higher requirements.
If one always drove with one hand on the side of the wheel, then NoA has zero impact on their driving style (no additional input required from you). If one drove hands off and only taped a button when nagged, then the new system does increase the number of button pushes in toto (assuming land changes are not synchronous to nag period).
Only one of these two scenarios falls within the operational guidelines of Tesla's system...

By more torque I mean over time, not in any one instance (though of course theoretically they could also increase the instance requirement but I have no reason to believe that).

If without ULC it takes ”sufficient torque applied every 10 seconds” (just to put a theoretical number on the frequency of the checks) and with ULC it takes ”sufficient torque applied every 10 seconds + sufficient torque applied at every lane change” the latter is more torque.

Certainly if you already apply more torque than is needed you don’t have to add more but the clearly not all do apply enough torque as it is...

Finally I am a bit disappointed in your examples. You clearly intentionally leave out the safest scenario of driving with both hands where the torque often fails to register even before ULC...

 

[mongo]

By more torque I mean over time, not in any one instance (though of course theoretically they could also increase the instance requirement but I have no reason to believe that).

If without ULC it takes ”sufficient torque applied every 10 seconds” (just to put a theoretical number on the frequency of the checks) and with ULC it takes ”sufficient torque applied every 10 seconds + sufficient torque applied at every lane change” the latter is more torque.

Certainly if you already apply more torque than is needed you don’t have to add more but the clearly not all do apply enough torque as it is...

Personally, I like the idea of using a torque bump to confirm change and direction.

Stepping back a bit, Tesla's system, in its current implementation, is meant to have the driver with hands on wheel and actively monitoring, thus the torque check should be continually satisfied. Any discussion of number of torque events operates outside the usage specification.

 

[electronblue]

Personally, I like the idea of using a torque bump to confirm change and direction.

Stepping back a bit, Tesla's system, in its current implementation, is meant to have the driver with hands on wheel and actively monitoring, thus the torque check should be continually satisfied. Any discussion of number of torque events operates outside the usage specification.

You know as well as I do that the torque check is not constantly satisfied when driving with both hands on the wheel.

 

[mongo]

You know as well as I do that the torque check is not constantly satisfied when driving with both hands on the wheel.

True, but that is a separate issue. If someone is dealing with alerts continuously, then offsetting the alerts by (max check time - minimum time for passed check before lane change) is not going to change the driving experience significantly.

Wasn't this side topic originally about whether the driver was confirming the lane change vs just a nag check (if nag was satisfied, no extra confirmation required)?

 

[electronblue]

True, but that is a separate issue. If someone is dealing with alerts continuously, then offsetting the alerts by (max check time - minimum time for passed check before lane change) is not going to change the driving experience significantly.

Wasn't this side topic originally about whether the driver was confirming the lane change vs just a nag check (if nag was satisfied, no extra confirmation required)?

Yes. I just disagree with you additional nags would not change the driving experience. I think they would add basically a confirmation to ULC for many having both hands on wheel.

But I would agree we do not know yet if ULC adds nags or if it amounts to a similar amount of nags than without ULC and it is merely a sort of coincidence they show up during ULC. Musk’s words leave the interpretation open though he does seem to suggest there is an additional check of some nature.

 

[ChrML]

Traffic laws do not apply in privately owned parking lots or driveways.

Actually this is wrong, atleast in Norway. I remember the traffic schools being very clear about the traffic laws applying everywhere.

Obviously people can't put up stop signs on their property and have the police fine people breaking it, but if you move the car 2 meters while drunk on your own driveway and the police see you, or drive irresponsibly so that eventually people nearby could be endangered, there will be trouble.

Meaning using Enhanced Summon on parking lots have the same legal questions as using it on any public road. Altough a bit different because you could argue that the parking lot is safer than the roads, because usually lower speeds involved.

 

[mongo]

Actually this is wrong, atleast in Norway. I remember the traffic schools being very clear about the traffic laws applying everywhere.

Obviously people can't put up stop signs on their property and have the police fine people breaking it, but if you move the car 2 meters while drunk on your own driveway and the police see you, or drive irresponsibly so that eventually people nearby could be endangered, there will be trouble.

Meaning using Enhanced Summon on parking lots have the same legal questions as using it on any public road. Altough a bit different because you could argue that the parking lot is safer than the roads, because usually lower speeds involved.

In the US (or at least Michigan), it depends if your driveway is publicly accessible...
Man can't be prosecuted for driving drunk in driveway, court rules

See also drag strips...

I do wonder how self driving cars will indicate to officers/ the public that the fact that the drier is unconscious/ not visible is ok...
Is that an FSD Tesla or not???

 

[ChrML]

In the US (or at least Michigan), it depends if your driveway is publicly accessible...
Man can't be prosecuted for driving drunk in driveway, court rules

Ah could be different then. Know a guy around here that got prosecuted for moving his car 3m drunk in his own driveway, and the police happened to walk by when he was done.

 

[S4WRXTTCS]

How many times do i have to be right?

This is what I assumed as well, but it actually turns out I was wrong.

How do I know? From actually using it.

It does not require any special turning force before it initiates the lane change. All it does is confirms that hands are detected on the steering wheel (using the torque sensor) before initiating the lane change.

In actual use it's seamless as long as you have your hands on the steering wheel in such a way that satisfies the torque sensor. This is even in the Model 3.

For all practical purposes I've concluded that I was wrong.

It doesn't require confirmation using the wheel. Instead the confirmation is assumed from the driver not canceling it within the 1-2 seconds before it turns on the turn signal.

Lesson Learned: Actually test things out even when things seem perfectly reasonably. I also learned that the Model 3 is more sensitive to hands on the steering wheel than I thought. I kinda wish they would simply add some kind of indicator to let me know (with little latency) if it's detecting my hands.

 

[electronblue]

This is what I assumed as well, but it actually turns out I was wrong.

How do I know? From actually using it.

It does not require any special turning force before it initiates the lane change. All it does is confirms that hands are detected on the steering wheel (using the torque sensor) before initiating the lane change.

In actual use it's seamless as long as you have your hands on the steering wheel in such a way that satisfies the torque sensor. This is even in the Model 3.

If you don't then you'll get nagged a bunch, and I believe ALL types of automatic lane changes won't work if hands aren't detected on the steering wheel. But, I need to confirm that.

For all practical purposes I've concluded that I was wrong.

It doesn't require confirmation using the wheel. Instead the confirmation is assumed from the driver not canceling it within the 1-2 seconds before it turns on the turn signal.

Lesson Learned: Actually test things out even when things seem perfectly reasonably. I also learned that the Model 3 is more sensitive to hands on the steering wheel than I thought. I kinda wish they would simply add some kind of indicator to let me know (with little latency) if it's detecting my hands.

Still not answering the question though: Will a person whose normal torque is not enough to dismiss nags (eg person with both hands on the wheel) get more nags or the same amout of nags with ULC (or even less)?

 

[S4WRXTTCS]

Still not answering the question though: Will a person whose normal torque is not enough to dismiss nags (eg person with both hands on the wheel) get more nags or the same amout of nags with ULC (or even less)?

I wasn't answering any questions.

All I was doing was confirming (after testing) that NoA did not require anymore torque to initiate a lane change than the torque required to dismiss/prevent nags.

I can't answer your question because I need to test one more thing.

I never tested auto-lane changes with ULC turned off to see if these also required hands to be detected on the steering wheel before initiating the lane change.

I've heard reports that it checks for this even though you've already used the stalk to confirm the lane change. But, I'll have to try it out because I've been under the assumption that it wasn't necessary. This assumption might be wrong because I've had issues in the past where auto lane changes were slow to start. So it might have been user error, and not the system being glitchy like I assumed.

In testing 2019.8.5 I didn't have any issues with the delay when lane changes started. They were all pretty consistent as long as a traffic situation didn't delay them. There was one time it canceled it because the car in front of me sped up. At first it was kinda "umm, why'd you do that", and then I saw the car ahead of me was further away.

I only had one nag in a round trip journey of 120 miles that mostly consisted of 80mph on a three lane freeway. Normally I have a lot more, but I was consciously making sure I was applying a little torque to the steering wheel.

To be honest regardless of ULC the whole steering torque thing is always going to be a thorny issue. It is because it's not the right sensor for the job. Tesla assumed they'd be rapidly going to L3 or above (where it won't be needed), but didn't consider that they'd be stuck on L2 for years. Now it seems like everything is getting tied to this annoying sensor. The worst is they don't have any real-time indication of the status of the sensor.

If I get really mad I'll replace it with a capacitive touch sensor. I figure I can mimic the output to match that of the torque sensor to trick it. It's still a detection device (unlike weights) so it's not a cheat device like the banned autopilot buddy.

 

[mongo]

Still not answering the question though: Will a person whose normal torque is not enough to dismiss nags (eg person with both hands on the wheel) get more nags or the same amout of nags with ULC (or even less)?

The person will exert less total torque than if they did the lane changes themselves.
Less movement than if they used the stalk to initiate the change too...

 

[NerdUno]

By the way, we got another Musk tweet for "Group 1" to be excited about and for "Group 2" to scoff at:

View attachment 394716

Seems Musk is promising that Autopilot will eventually be able to handle potholes.

Yeah. Can't wait to see where the car goes when it spots a pothole... 18-wheelers beware.

 

[mhan00]

Maybe we are talking in circles.

Let me explain this to you again. Periodically, the car checks to see if the driver has their hands on the wheel by checking for torque. Before every lane change, the car does an extra check to see if the driver still has their hands on the wheel by checking for torque. If it senses enough toque, then the driver passes the check and the car will then proceed with the auto lane change without any additional driver input. If the driver fails the check (not enough torque), then the car will prompt the driver to tug the wheel to confirm that they are holding the wheel before executing the auto lane change. So if the driver passes the check, the auto lane change happens with no additional driver confirmation. If the driver fails the check, then in that case, the car will require driver confirmation before executing the lane change. So the car does not require a confirmation for every single lane change as you claim, it only requires it if you fail the check.

That is what Musk is saying in that tweet and what multiple users have confirmed.

I can confirm it in my case too. Since I always have the wheel and do provide a little torque nearly constantly, I have had multiple moments where I got the lane change alert and simultaneously had my blinker turn on without me consciously confirming anything.

 

[diplomat33]

Some more info on the AP3 computer:

- The Tesla Full Self-Driving Computer (AP3) now in production is at about 5% compute load for these tasks [Nav on AP] or 10% with full fail-over redundancy. For comparison, the current AP2.5 is at 80% compute load on Nav on AP.
- AP3 redundancy "Two, independent system-on-chip architecture, with each SoC having two NN accelerators that can perform simultaneous health-check calculations to protect against a soft error.”"
https://electrek.co/2019/04/09/tesla-autopilot-computer-5-capacity-current-80/

It would seem AP3 should give Tesla a lot more computational power for expanding what Autopilot can do.

 

[CarlK]

Some more info on the AP3 computer:

- The Tesla Full Self-Driving Computer (AP3) now in production is at about 5% compute load for these tasks [Nav on AP] or 10% with full fail-over redundancy. For comparison, the current AP2.5 is at 80% compute load on Nav on AP.
- AP3 redundancy "Two, independent system-on-chip architecture, with each SoC having two NN accelerators that can perform simultaneous health-check calculations to protect against a soft error.”"
Tesla’s new Autopilot computer is only at 5% capacity, current 2.5 at ~80%

It would seem AP3 should give Tesla a lot more computational power for expanding what Autopilot can do.

Good to hear that. Many people were driveling on camera, Lidar or whatever sensors but the challenge has always been the brain power as Elon correctly saw from the begining. The machine "brain" consists of the neuron equivalence of AI processor and synapsys equavalence of NN machine learning firmware. Tesla is leading in both fronts because it has focused on the basics instead of superficial things. Another win for the first principle of engineering if Tesla becomes the first company to come out with the autonomous technology which I think it will be.

 

[stretchsje]

So, four NN chips on the FSD computer. Not two. If the earlier estimates of TOPS per NN processor are correct, this isn't that far off from nVidia's estimates for computational power required for L5 (estimated 160-240 for Tesla not including whatever small boost the Samsung chip provides, 320 total for nVidia). I'm going to wager that the Tesla chips operate closer to their theoretical throughput due to a more optimized memory system.

 

[CarlK]

So, four NN chips on the FSD computer. Not two. If the earlier estimates of TOPS per NN processor are correct, this isn't that far off from nVidia's estimates for computational power required for L5 (estimated 160-240 for Tesla not including whatever small boost the Samsung chip provides, 320 total for nVidia). I'm going to wager that the Tesla chips operate closer to their theoretical throughput due to a more optimized memory system.

And the Tesla AI chip is tailored for its own NN algorithm. That's the advantage of vertical integration and reason why Elon wants to go that direction for everything he does even it's a much harder way to go. Those "consortiums" just don't stand a chance competing.

 

[electronblue]

Some more info on the AP3 computer:

- The Tesla Full Self-Driving Computer (AP3) now in production is at about 5% compute load for these tasks [Nav on AP] or 10% with full fail-over redundancy. For comparison, the current AP2.5 is at 80% compute load on Nav on AP.
- AP3 redundancy "Two, independent system-on-chip architecture, with each SoC having two NN accelerators that can perform simultaneous health-check calculations to protect against a soft error.”"
https://electrek.co/2019/04/09/tesla-autopilot-computer-5-capacity-current-80/

It would seem AP3 should give Tesla a lot more computational power for expanding what Autopilot can do.

I envy you @diplomat33.

I really do.

There was a time when I also felt that naive excitement about all things Tesla and all things Tesla Autopilot. I just can’t feel it anymore.

Good to hear that. Many people were driveling on camera, Lidar or whatever sensors but the challenge has always been the brain power as Elon correctly saw from the begining. The machine "brain" consists of the neuron equivalence of AI processor and synapsys equavalence of NN machine learning firmware. Tesla is leading in both fronts because it has focused on the basics instead of superficial things. Another win for the first principle of engineering if Tesla becomes the first company to come out with the autonomous technology which I think it will be.

That is certainly the 2016 thesis. How much water it still holds in 2019 remains to be seen.

 

[electronblue]

That B pillar camera eye.

I’m pretty sure it mocks me every time I approach my car.

”Ha-ha! Fool!”